A step-by-step guide to building a key-value database from scratch: https://t.co/DqJ6fI0knU - love the explainer/interactive animation in this

Oooh @yworks is niiiiice!

Is Cytoscape still the best tool for making really nice property graph viz? I used it 10+ years ago, just wondering if tooling has moved on to anything more user-friendly for n00bs in the field :)

A fun few hours with Cursor, neo4j, and Cytoscape analysing astroturfing and sock-puppet accounts on Reddit. Common pattern of a handful of accounts posting 'questions', and then another comments on it. Maybe an attempt at seeding LLMs, which scrape Reddit?

Colocating Input Partitions with Kafka Streams When Consuming Multiple Topics: Sub-Topology Matters! - Vishal Sharma

🌶️ Hot take: if you have to label your opinion as a 'hot take'… it probably isn't.

My theory: Building LLMs and even inference ecosystem requires an organization with very strong data muscles. The ability to ingest, clean up and process gigantic amounts of data effectively. Google, thanks to its ads-based business, was always a leader in data use. Inventing

When OpenAI released ChatGPT, many people said that Google and Apple lost the AI race. 3 years later, it’s clear that Google recovered and caught up. While Apple seems close to giving up. What do you think caused the difference? (I’ll share my own theory in a bit).

In the eternal struggle between Good vs Evil, Blur vs Oasis, and @duckdb vs @ApacheDataFusio , we just switched to DataFusion after 18 months, while keeping our #FaaS magic intact: https://t.co/wF2GsLkPHq

good stuff from @vanlightly: How Would You Like Your Iceberg Sir? Stream or Batch Ordered? https://t.co/0pVE6X91DL

One for my DBA friends! A very good talk from @andy_pavlo, looking at how LLMs can _actually_ be used to help improve performance. It's measured and reasonable, with plenty of caveats and real-world considerations from someone who _really_ gets RDBMS. https://t.co/aiIhQZ8Iz0

I missed this release last week - #ApacheFluss now supports writing to both #ApacheIceberg and Lance (as well as the original #ApachePaimon. https://t.co/qJtqSZYmSV

Protobuf - use it faster, or don't use it at all? Couple of interesting talk from P99 CONF this year: * https://t.co/uGWLcGTUdF * https://t.co/aHZ31dmq5N

Three Flink-related talks at P99 CONF this year - here are the recordings and slides: * https://t.co/sfI7LerNly * https://t.co/7yNHqhDZLD * https://t.co/aBwquLzRSS

P99 CONF nails it in terms of both content, *and* UX for attending. Super-simple & free registration. Videos available on-demand afterwards with no gating or games. Clear and usable website. These folk know how to do a developer conference. A+++ https://t.co/CNvMmGdfK7

How I create presentations with AI assistance. tl;dr: I find AI great for brainstorming, generating examples, and some research. Sometimes image generation works too. I can’t stand its writing style (and haven’t been able to get it to improve) and it cannot build entire talks

You can spot an aggrieved Linux loser by the attempt to gatekeep the term "distribution" like it was a royal distinction of honor. I don't give a fuck what you call a compilation of configs, tools, and programs with a custom installer that ships on an ISO.

Alex Jacobs: The Case Against pgvector

You Should Write An Agent

Having multiple copies of your data, laid out for different access patterns, is completely fine. Desirable even. Just make sure to have one canonical source of truth and drive updates to the copies from there.